Noon Deals Bot

Channel: t.me/noon_hot_deals · Site: ديلز مصر — vette1123.github.io/noon-deals-bot

Auto-posts the best discounted products from noon.com Egypt to a Telegram channel, 6× a day — fully free to run, no login required.

• Scrapes noon's deal pages using curl_cffi (Chrome TLS impersonation — no paid API)
• Attaches your influencer coupon code to every post as tap-to-copy text
• Posts Arabic-formatted product cards with images to Telegram
• Republishes the same deals as a static Arabic site on GitHub Pages so search traffic earns too
• Optionally crossposts to a Facebook Page
• Runs on GitHub Actions — no server required

How it works#

GitHub Actions (cron, 6×/day)
        │
        ▼
   fetch 6 category feeds  ── curl_cffi + Chrome fingerprint ──▶  noon.com
        │                     (rotating ~34 categories × 10 pages)
        ▼
   filter, then rank by expected commission in EGP
        │
        ▼
   post to Telegram channel with:
     • product image
     • Arabic-formatted card
     • tap-to-copy coupon code (e.g. gado)
     • "Buy now" button → noon.com product URL + affiliate UTMs
        │
        ▼
   commit updated state.json / posted.json / deals.json
        │
        ├──▶  crosspost to a Facebook Page (only if secrets are set)
        │
        ▼
   rebuild the static site from deals.json ──▶ GitHub Pages

Attribution model: two channels, both message-side, neither needs a login. Every "Buy now" link carries the affiliate campaign UTMs, and the message body shows a tap-to-copy coupon code that users paste at noon's checkout.

What gets posted: at least 25% off, at least EGP 150, not rated below 3.5 by 20+ buyers, and not unbranded-and-unreviewed filler. Survivors are ranked by expected commission in pounds — min(cap, category rate × price) times conversion odds — deduplicated by name, capped at 2 per seller. The top 12 go to Telegram; the top 20 are archived for the site.

Rates are per category and payout is capped at AED 50 per item, which makes expensive electronics (laptops 3%, mobiles 2%) the worst thing on the board and mid-priced beauty, toys and small appliances (8%) the best. See docs/MONETIZATION.md for the full table.

Feed cursor: noon's category browse paths are real product lists; each run walks 6 of them and advances through a rotation of ~34 categories × 10 pages (~14,000 products, roughly a 10-day cycle). A SKU already posted is on cooldown for 21 days.

The static site#

Every deal is archived to deals.json and rendered into an Arabic (RTL) site: a front page, one page per deal with Product structured data, a hub per category and per brand, a directory of each, a paginated archive, sitemap.xml, robots.txt and an RSS feed. No web fonts, no CSS files, no script bundles — search traffic arrives on mobile data, and page speed is ranking. It follows the device's light/dark setting, with a toggle that remembers an explicit choice.

It matters because it earns without an audience: a Telegram post reaches today's subscribers, a page that ranks reaches everyone who searches.

Two rules it will not break:

• Deal pages never 404. A deal that ages out becomes a noindex tombstone linking on to its hubs, because a URL that took four months to rank is the one thing here that cannot be rebuilt.
• priceValidUntil comes from when the deal was seen, not from the build, so a rebuild never re-certifies an old price as current.

Enable it once under Settings → Pages → Source: GitHub Actions; the publish-site job does the rest. To move it onto your own domain, set the SITE_DOMAIN repository variable and point the DNS at Pages — worth doing before it ranks, not after.

Getting indexed#

Every build re-emits two ownership files at the site root, and every deploy announces what changed:

• Google Search Console — google<token>.html, served verbatim. Verify the property once, then submit sitemap.xml. A fork points GOOGLE_SITE_VERIFICATION at its own token.
• IndexNow — <key>.txt plus a POST to api.indexnow.org after the deploy, listing only the pages that run changed: the front page, the new deal pages and the hubs they landed on. Bing, Yandex and Seznam then crawl in minutes instead of weeks. Google does not participate — there, sitemap.xml + lastmod is still the lever. The key is public by design (serving it is the proof of ownership), so it belongs in INDEXNOW_KEY as a variable, never a secret.

Neither can fail a run: a malformed token is dropped with a warning, and a rejected ping is logged and ignored.

Preview it locally:

python site_builder.py && python -m http.server -d public 8000

See docs/MONETIZATION.md for what earns, what is built, and what still needs a human.

Project layout#

Local setup#

pip install -r requirements.txt
cp .env.example .env  # fill in values
python main.py --dry-run   # prints what would be posted, skips Telegram
python main.py             # real run

Since mid-2026 Akamai also challenges cold requests from residential IPs, so a local run may fail at the fetch step with Akamai JS challenge. CI is unaffected (it egresses through Cloudflare WARP). To debug parsing locally, save the page HTML from a real browser and feed it to scraper.parse_products_from_html.

Required secrets (GitHub Actions)#

Two are truly required. No scraping API key, no noon.partners login, no session rotation, no OTP flow.

Running tests#

pip install -r requirements-dev.txt
pytest

Schedule#

Cron 0 4,8,12,16,20,0 * * * UTC → 6×/day (every 4 hours). Triggered by .github/workflows/bot.yml or manually via workflow_dispatch.